Sebastian Kosch
Sebastian Kosch
> given that we have autospacers already I mean, the output of your net already includes whatever sidebearings you started out with, so in a way you're very much already...
(P.S. I hope I will find some time soon to play with your code myself. Talk is cheap, I know!)
> The problem with looking at non-kerned pairs is that you never know if the designer deliberately left גT unkerned because it already fits just fine, or if there's no...
> it should work to pull spacing data out of this thing as well. True, for sure! What I mean though is that the horizontally-scanned contour samples fail to capture...
Really interesting plots, thanks for sharing! > Each of these inputs is chewed on separately by a convolutional layer, and then the results are concatenated together. Makes sense. However, you...
Nice! By the way I just came across an a relevant article today – they used restricted Boltzmann machine networks pre-trained on natural images, which is a pretty cool idea....